Re: OpenSSL build error on 8.2

[Romain Dolbeau <romain%dolbeau.org@localhost>]

Le lun. 18 mai 2020 à 12:39, Erik Fair <fair%netbsd.org@localhost> a écrit :
> It would be interesting to performance test the gcc -mcpu and -mtune sparc specific options on ssh and see if it’s better

For openssl, I have a opensl-1.1.1e-sparcv8 about to be uploaded to
dropbox, great minds think alike :-)

> - IIRC, there are some instruction additions to supersparc and hypersparc (both of which are documented options in gcc(1)) which might help.

v8 (implemented by SuperSPARC, HyperSPARC, MicroSPARC, but not the
original SPARC as seen in pre-10 SPARCstation which is v7) added the
integer multiplication in hardware, which helps a lot some algorithms.
That's why you want v8-compiled crypto on those CPUs.

v8+/v9 (UltraSPARC and later) also added compare-and-swap (CAS, also
some other less significant changes), which is nearly indispensable
for modern code that rely heavily on atomic operations. The lack of
CAS is the primary reason why pre-v8+/v9 hardware is utterly obsolete
today, in my opinion (unfortunately, as obviously I'm a SPARC fan :-)
).

Cordially,

-- 
Romain Dolbeau